a distributed optimized patient scheduling using partial information

7/31/2019 A Distributed Optimized Patient Scheduling using Partial Information

1/12

International Journal of Artificial Intelligence & Applications (IJAIA), Vol.3, No.3, May 2012

DOI : 10.5121/ijaia.2012.3307 83

A Distributed Optimized Patient Scheduling usingPartial Information

G. Mageshwari and E. Grace Mary Kanaga

Department of Computer Science and Engineering, Karunya University, Coimbatore,

[email protected]

[email protected]

ABSTRACT

A software agent may be a member of a Multi-Agent System (MAS) which is collectively performing a

range of complex and intelligent tasks. In the hospital, scheduling decisions are finding difficult to schedulebecause of the dynamic changes and distribution. In order to face this problem with dynamic changes in the

hospital, a new method, Distributed Optimized Patient Scheduling with Grouping (DOPSG) has been

proposed. The goal of this method is that there is no necessity for knowing patient agents information

globally. With minimal information this method works effectively. Scheduling problem can be solved for

multiple departments in the hospital. Patient agents have been scheduled to the resource agent based on the

patient priority to reduce the waiting time of patient agent and to reduce idle time of resources.

KEYWORDS

Patient Scheduling, Multi-Agent System, Software Agent, Distributed Computing.

1. INTRODUCTION

There are many kinds of software agents, with different characteristics such as mobility,

autonomy, collaboration, persistence and intelligence. Software agents are used to solvedistributed problem. An agent is a software entity or computer system that is capable of

autonomous action in order to meet its goal [1]. Some of the characteristics of agent arereactivity, proactiveness and social ability. Reactivity means agents are able to respond to thetimely changes that occur in order to satisfy its design goal. Proactiveness means agents are able

to exhibit the behavior by taking initiative in order to satisfy its design goal. Social ability, thischaracteristic tells that agents are capable of interacting with each other agent in order to satisfy

its design goal. Some additional characteristics of agents are adaptability, autonomy,collaboration, knowledgeable, mobility and persistence [2].

There are different types of agents among which personal agents, mobile agents, collaborative

agents are provided the characteristics such as mobility, collaboration, intelligence or flexibleuser interaction. Personal agents interact directly with a user, presenting some personality orcharacter, monitoring and adapting to the user's activities, learning the user's style andpreferences, and automating or simplifying certain rote tasks Multiple software agents are suitable

for use in a wide variety of application that involve in distributed computation to solve complexproblems or communication between the components. Mobile agents are collecting information

or perform actions when it has been sent to remote sites. Collaborative agents are communicating
mailto:[email protected]:[email protected]:[email protected]:[email protected]


2/12


84

with groups, representing users, organizations and services. Multiple agents share the messages tointeract with each other. MAS change the complex problem into simple which can be solved

easily. MAS is a system composed of multiple interacting intelligent agents. It can be used tosolve problems that are difficult or impossible for an individual agent.

Agent abilities are collecting data from numerous places, searching and filtering, monitoring,agent to agent negotiation and parallel computations. The agent negotiates with resource usage.Each agent owns set of tasks, requiring certain resources for a given processing time. Software

agents are in the following applications such as Personal Information Management (PIM),Electronic Commerce (EC) and Business Process Management (BPM) [3]. Multi-agent systemscooperate with other agents to solve more complex tasks. Ability of multi-agent is for increasing

the quality of distributed computing. Complex software systems are still difficult to implement.Simulation techniques for MAS can be utilized to test and understand the runtime behaviour of a

distributed time-based system [4]. Complex problems can be changed in to individual pieceswhich can be easier to implement.

2. STATE OF ART

Dynamic load balancing considers computational load and communicational to decrease theexecution time and to increase the resource utilization. In this method, monitoring phase,redistribution phase, migration phase are considered by Robson E.De Grande et al [5]. Firstphase is detecting load imbalance using data collection, filtering and selection of data. Second

phase determine possible migration moves. Then transfer from an overloaded resources to anunder loaded resources. Computational load imbalance is avoided when the load is redistributed

for communication purpose. Third phase consider two phase migration technique. It is used in

redistribution scheme to minimize the migration latency and enable better balancing reactivity.According to Feng Zhang et al [6], price based multi-agent scheduling and coordination is used to

determine the timeliness of scheduling in a distributed environment. This framework applied inthe inventory system. It is scalable. The number of asset is varied to reflect the different problem

scales. But the agents cannot retrieve the information automatically. Intra-organization as well asinter-organization coordination needs to be improved. Decision process is used for non-emergency admission planning in the hospital activities.

According to Luiz Guilherme Nadal Nunes et al [7], Markov Decision Process (MDP) is used to

guide more efficient decision process to balance the non-emergency patients admission andavailable resource utilization in the hospital. Medical Path Coordination (MedPaCo) is a

coordination mechanism which is used to minimize waiting time of the patients [8]. The idea of

this coordination mechanism is that the patient agents are buying the time slots of resources forthe needful treatments and examinations of the patients. Adaptive dynamic approach is used to

schedule the resources automatically. It minimizes the access time of all the resources in thehospital. According to Ivan B.Vermeulen et al [9], adaptive urgent scheduling schedules the

emergency patients on time. It provides flexible resource allocation for emergency patients.Health Examination Scheduling Algorithm (HESA) is used to allot the examinee to any freeresources or doctor based on round robin algorithm. This method assigns the priority for the

patient based on their treatment. But it does not consider, when the examinees are absent,equipment breakdown or longer than expected processing times in the resources [10]. Multi-agent

Pareto Appointment Exchanging (MPAEX) system is used to exchange appointment schedulessuch that no patient is worse off. It uses negotiation approach to improve the condition of patient

which is called Pareto improvement [11], [12]. Partial plan is developed for the patient with less

number of tasks or activities. It can exchange single activity. Workload distribution has beenscheduled using theil index in MPAEX approach. Generalized Partial Global Planning (GPGP)


3/12


85

with new resource constrained mechanism is used to improve the throughput of the hospital unitand reduces the stay time of the patient [13]. Bidding is used in GPGP coordination mechanism

to buy the resource timeslot for the patient agent. When two tasks need to use the same resourceat a time, highest priority will be given to the task which is arrived the hospital first. Newcoordination mechanism is introduced with GPGP approach for handling mutually exclusive

resource relationship [14].

Dynamic changes of patient scheduling problem is solved by agent based, patient-centredapproach [15]. This approach is used to avoid the resource conflicts. Medical Path Agent

(MedPAge) shows the roles of the agent, dynamic interaction and the coordination object

identification [16]. MedPAge project handles the complicated tasks through the marketmechanism which is supported by software agent in the hospital. To improve the current

scheduling situation, agent interaction protocol is used. The agents interact with each other. Thismechanism maximizes the patients own utility. Distributed multi-agent based approach isimplemented in the Medical Path Coordination (MedPaCo) for coordination of patient agents

[17]. The aim of this coordination mechanism is to reduce the overall waiting time of the patients.This approach implements the patients and hospital resources as software agents through the

concept of MedPaCo [18]. This mechanism is based on auction or bidding process which consists

of certain phases. These phases make the patient to get highest utility from a timeslot by payinghighest amount for that timeslot. This coordination mechanism cannot handle multiple

preferences of the patient agents.

3. PROBLEM DESCRIPTION

Scheduling is difficult to implement in the hospital because of the dynamic changes. Medical datamanagement systems and Decision support systems are analyzed by agent based approach.

Medical data management systems are focused on the processing of medical data and retrieval ofdata such as electronic health record. Decision support systems approaches are aimed to help the

execution of healthcare processes such as treatments [4]. According to ReddyMC et al. [19], [20]

challenges are found in the hospital from various analysis. First challenge is ineffectiveness ofcurrent information and communication technologies. Cell phones, 2-way radios are the

communication technologies which is failed to work effectively because of mismatch frequencies.Clinical/nonclinical department interaction makes inappropriate patient transfer. Clinical/clinical

department interaction increases the waiting time of the patient in a department. Lack of completeand accurate information shared between the departments because of dynamic changes in the

hospital. Without appropriate information, inpatient access department may make inappropriateassignments as well as deal with temporary bottleneck problem in the patient flow by holding upwith resources. Breakdowns in information flow include the loss of patient information

transferring the patients to wrong locations due to misrepresentation of patients information .Master health checkup includes blood test, urine test along with Electro cardio gram (E.C.G),

ultra scan, X-ray, thyroid function test, dental consultation and eye consultation. In master health

checkup, patients can undergo any checkup. When a particular resource gets overcrowded thepatients can go any free resources because there is no constraint. New method has been proposed

in this paper to transfer the group of patients from overcrowded resources to free resources.

4. THE PROPOSED DISTRIBUTED PATIENT SCHEDULING WITH GROUPING

METHOD

This section describes the architecture of patient scheduling, flow chart and the algorithm of theproposed method.


4/12


86

4.1. Architecture Overview

Agent scheduling process is developed by Qingqi Long et al [21] for dynamic load balancing inagent based simulation. Each agent executes the simulation by message communication. Based on

this agent scheduling, this paper propose a method to schedule the patients to the resources in

the hospital. The architecture of patient scheduling is shown in Figure 4.1.

Figure 4.1 Architecture of Patient Scheduling

Master Health Checkup Agent (MHCA) is in charge of receiving, processing the Patient Agent

(PA) information. It also maintains the count of patients. Resource Agents (RA) send and receivethe requests from other resources. Responsibility of MHCA is to send a PA to the RA. It is also incharge of monitoring PA and RA information and agent communication. In addition, each

resource is equipped with a Scheduling Agent (SA) that is in charge of triggering patientscheduling procedure, selecting the migrated agent sets and target containers, and moving the

agents. The SA is the centre of patient scheduling and migration in this simulation platform. Aseach resource has a SA, the scheduling is done in a distributed way.

When a paricular resource gets overcowded it needs to migrate the exceeding number of patientsto another resource in order to reduce the waiting time of the patient and to improve the resource

utilization. First the overcrowded resource will send request to neighborhood resource which

response if it has enough space to treat that patient. Otherwise it will ignore the request. After thatovercrowded resource will send request to another resource until overcrowded will be accepted

by any neighborhood resource. Using migration set approach, minimizes the number ofcommunication and grouping the patient agent based on their treatment. Resource agents will

interact with each other. Each resource agent does not have any information about the patientagent in another resource agent. Figure 4.2 show that how a set of patients migrated fromovercrowded resource agent to neighbourhood resource agent.


5/12


87

Figure 4.2 Set of Patient Transferred using Migration Set Approach

4.2. Algorithm and Flow Diagram of Proposed Method

Table 4.1 shows the notations which are used in this method. The proposed method reduces the

waiting time of the patient and improves the patient satisfaction.

Table 4.1 Notations


6/12


88

Figure 4.3 shows the flow diagram of transferring set of patients. represents the overcrowded

resource agent. represents the neighborhood resource agent without crowd. When a

particular resource agent gets overcrowded then that will send request to another

neighborhood resource agent. It will check whether it has enough capacity to do treatment for

another set of patient. If it has space it will accept. Otherwise it will reject the request fromovercrowded resource agent. Group the patients from exceeded number of patient based on their

priority, new arrival time and next task. That group of patients will be considered as a set

.

Figure 4.3 Flow Diagram for Transferring Group of Patients

Algorithm1: Distributed Optimized Patient Scheduling using Partial Information

sends request toIf( < ) then

Accept request

ElseReject request

End if

For ( ) do

Group the set (i.e) set of patient in exceeded number of patientMove to

Remove fromEnd for

Figure 4.4 Distributed Optimized Patient Scheduling using Partial Information

Send request to neighborhood resource agent

Receive the request from overcrowded resource agent

If (

)

Group the patient in overcrowded resource

agent

Transfer a set of patient from overcrowded

resource agent to neighbourhood resource agent

NO

YES

Start

Stop


7/12


89

The pseudo-code of the proposed algorithm is given in Figure 4.4. When a particular resourceagent gets overcrowded then that will send request to another neighborhood resource agent. It will

check whether it has enough capacity to do treatment for another set of patient. If it has space itwill accept. Otherwise it will reject the request from overcrowded resource agent. Group thepatients from exceeded number of patient in the overcrowded resource agent based on their

priority, new arrival time and next task. That group of patients will be considered as a set .

Finally, patients are transferring from overcrowded resource agent to neighborhood resource

agent. Set of patients will be moved to another resource agent. Then migrated agent will be

removed from overcrowded resource agent. It will be repeated until the entire overcrowded

patient is scheduled. This method DOPSG has been used to search for optimized schedulingscheme, including migrated agent sets.

The steps of the proposed algorithm are as follows,

Step1: send request to with id of and exceeded number of patients ( ).

Step2: If ( < ) then accept the request. Otherwise go to step1.

Step 3: When a patient belongs to exceeded number of patient then that patient has to be grouped

based on priority, new arrival time, and next task. Group of patients considered as a set .

Step4: Move to and remove from .Step5: Repeat until all exceeded patients transfer from overcrowded resource agent.

Step 6: Stop.

5. EXPERIMENTAL RESULTS

To implement this scheduling method for distributed patients with grouping, hardwarerequirements and software requirements are used. Hardware Requirements are Intel Core (TM)

2 Duo CPU T8100 @ 2.10 GHz processor and 32- bit operating system are used. SoftwareRequirements are JDK 1.6 java software used with Java Agent Development Framework (JADE)

package. Three resources and fifty patients are created to do the scheduling. Initially create thepatient agent in MHCAgent by retrieving from database. Each patient agent have unique agent idwhich will be generated while creating agent. Created patient agent should send to resource agent

to do their treatment. Through agent communication one resource agent sends the message toanother resource agent. After receiving the responses, group of patient agents are moved from one

resource agent to other resource agent. It minimizes the number of communication. Then

scheduling has been done with grouping.

5.1. Performance Metrics

The performance of the multi-agent system can be evaluated by following metrics. The aim ofpatient scheduling problem is to minimize the patient waiting time and idle time of resources in

the hospitals. Performance metrics are useful in evaluating the objectives of patient scheduling isgiven below:

Maximum completion time (Cmax)

This refers to difference between the starting time and finishing time of a sequence of jobs ortasks. In terms of patient scheduling, it refers to the difference between when the patient checks

in and the time taken to complete all the tasks in their diagnosis.

Cmax= Maximum of end time


8/12


90

Maximum tardiness (Tmax)

It refers to the lateness of a job which is a value equal to the completion time minus the due time.This metric gives the worst case scenario of how long a patient might be delayed in the worstcase.

Tmax= Maximum of TardinessTardiness=Cmax DuetimeDuetime=Arrivaltime+Totalprocessing time

Total completion time (Cj)

It represents the total amount of time utilized in the schedule. For patient scheduling we want to

minimize this metric to minimize the patient stay in the hospital.

Cj= sum of completion time

Total tardiness (Tj)

It measures the overall total tardiness of all the patients.

Tj= sum of tardiness

Total weighted completion time (WjCj)

This metric includes the priority of the patient when calculating the total flow time.

WjCj= sum (weight*completiontime)

Total weighted tardiness (WjTj)

This is the main metric is used to evaluate the patient scheduling algorithm to indicate whether

the patient waiting time is minimized.

WjTj= sum (weight*tardiness)

A good scheduling technique will effectively minimize the values of all these metrics. Thisproposed approach also aims to minimize the weighted tardiness and the weighted completion

time.

5.2. Performance Analysis

Performance analysis has been shown in the graph. Figure 5.1 shows the comparison of total

completion time of the patients of proposed method with traditional scheduling techniques likeFirst Come First Serve (FCFS), Weighted Shortest Processing Time (WSPT), Distributed

Optimized Patient Scheduling (DOPS) and Distributed Optimized Patient Scheduling withGrouping (DOPSG).


9/12


91

Figure 5.1 Comparison of Total Completion Time

DOPSG minimizes number of communication than the other scheduling. The total completiontime of DOPSG is reduced by 1% when compared to DOPS, 1.7% when compared to WSPT and

1.7% with respect to FCFS. Figure 5.2 shows total tardiness of the patient in various schedulingtechniques. The total tardiness of DOPSG is reduced by 9% when compared to DOPS, 12% whencompared to WSPT and 12% with respect to FCFS.

Figure 5.2 Comparison of Total Tardiness

Figure 5.3 shows the total weighted completion time of the patients in various schedulingapproaches. The total weighted completion time of DOPSG is reduced by 0.8% when compared

to DOPS, 1.1% when compared to WSPT and 1.2% with respect to FCFS.


10/12


92

Figure 5.3 Comparison of Total Weighted Completion Time

Figure 5.4 shows total weighted tardiness of the patient in various scheduling methods. The total

weighted tardiness of DOPSG is reduced by 5% when compared to DOPS, 37% when comparedto WSPT and 41% with respect to FCFS.

Figure 5.4 Comparison of Total Weighted Tardiness

6. CONCLUSIONS

A distributed approximate optimized patient scheduling algorithm with partial information hasbeen proposed and implemented in this paper. First, this algorithm does not need to know all the

information of departments where agents are, so scheduling can be carried out with partial


11/12


93

information. Secondly, since this algorithm is distributed, it can not only avoid bottleneckproblem and potential failure problem caused by the centralized mode, but also improve resource

utilization. Thirdly, this algorithm is approximately optimized. This proposed algorithm is used toreduce the waiting time of patients and thus improving the health state of patient and to improveresource utilization.

REFERENCES

[1]. Michael Wooldridge, An Introduction to Multi agent systems 2nd edition, Wiley, 2009.

[2]. Martin L. Griss and Edited George and T. Heineman and Ph. D and Ph. D and William Councill,

Software Agents as Next Generation Software Components, In Component-Based SoftwareEngineering, Addison-Wesley ,pp.641-657, 2001.

[3]. Nick Jennings, Michael Wooldridge, Software Agents, IEE Review, pp 17-20 Jan 1996.

[4]. David Isern, David Snchez, Antonio Moreno, Agents applied in health care: A review internationaljournal of medical informatics vol.7 9, pp. 145166,2010.

[5]. Robson E.De Grande, Azzedine Boukerche, Dynamic balancing of communication and computaionload for HLA based simulations on large scale distributed systems, in proceedings of the EEE

/ACM international symposium on distributed simulation and real time applications,pp.175-183,2009.

[6]. Feng Zhang, Eugene Santos, Jr., and Peter B. Luh, A Performance Study on a Multi-agent E-Scheduling and Coordination Framework for Maintenance Networks, IEEE Transactions on

systems, man, and cyberneticspart c: applications and reviews, vol. 37,pp. 52-65,2007.

[7]. Luiz Guilherme Nadal Nunes, Solon Venncio de Carvalho, Rita de Cssia Meneses Rodrigues

Markov decision process applied to the control of hospital elective admissions, Artificial

Intelligence in Medicine,Volume 47, Issue 2, Pages 159-171, October 2009.

[8]. T. O. Paulussen, A. Zller, A. Heinzl,A. Pokahr, L. Braubach, W. Lamersdorf,Dynamic PatientScheduling in Hospitals, In Agent Technology in Business Applications (ATeBA-04), 2004.

[9]. Ivan B.Vermeulen,Sandar M.Bohte, Sylvia G.Elkhuizen,Han Lameris, Piet J.M.Bakker, Han La

Poutre, Adaptive resource allocation for efficient patient scheduling, Artificial Intelligence inMedicine, vol. 46,pp. 67-80, 2009.

[10]. Ching-Chin Chern ,Pei-Szu Chien , Shu-Yi Chen, A heuristic algorithm for the hospital healthexamination scheduling problem, European Journal of Operational Research,pp.1137-1157,2008.

[11]. Vermeulen I, Bohte S, Somefun K, La Poutr H, Improving patient activity schedules by multi-agentpareto appointment exchanging , in Proceedings of the IEEE international conference on E -commerce technology, CEC/EEE, pp 5663,2007.

[12]. Ivan Vermeulen, Sander Bohte, Koye Somefun, Han La Poutr, Multi-agent Pareto appointmentexchanging in hospital patient scheduling, Springer, SOCA 1:185196,2007

[13]. Decker, Jinjiang Li , Coordinated hospital patient scheduling, IEEE International Conference onMulti agent system, pp. 104-111,1998.

[14]. Keith Decker and Jinjiang Li, Coordinating Mutually Exclusive Resources using GPGP,Autonomous Agents And Multi-Agent Systems,Volume 3, Number 2, 133-

157, DOI: 10.1023/A:1010074611407,2000.[15]. T.O. Paulussen, N.R. Jennings, K.S. Decker, and A.Heinzl, Distributed patient scheduling in

hospital,International Joint Conference on Artificial Intelligence,vol. 18, pp. 1224-1232, 2003.

[16]. A. Bartelt and W. Lamersdorf and T. O. Paulussen and A. Heinzl, Agent Oriented Specification forPatient-Scheduling Systems in Hospitals,in Dagstuhl Article,2002.


12/12


94

[17].1Torsten Paulussen,Anja Zller,Armin Heinzl,Alexander Pokahr, Lars Braubach, Winfried Lamersdor

f,Dynamic Patient Scheduling in Hospitals , Coordination and Agent Technology in ValueNetworks,2004.

[18]. T.O Paulussen, A Zller, A Heinzl, L Braubach, A Pokahr , W Lamersdorf, Agent-based patient

scheduling in hospitals, In S Kirn, O Herzog, P C Lockemann and OSpaniol, eds., MultiagentEngineering Theory and Applications in Enterprises, Springer Verlag, Heidelberg, Germany, pp.

255-276,2006.

[19]. ReddyMC, Paul SA,Abraham J,MCNeese M,DeFlitch C, Yen J, Challenges to effective crisismanagement: using information and communication technologies to coordinate emergency medical

services and emergency department teams, International Journal of medical informatics, vol. 78, pp.259-269,2009.

[20]. Joanna Abraham, Madhu C.Reddy, Challenges to inter-departmental coordination of patienttransfers:A workflow perspective, International Journal of medical informatics, vol. 79, pp. 112-122,2010.

[21]. Qingqi Long , Jie Lin, Zhixun Sun, Agent scheduling model for adaptive dynamic load balancing inagent-based distributed simulations, Journal of Simulation Modelling Practice and Theory, Vol.19,pp. 10211034, 2011.

a distributed optimized patient scheduling using partial information

Documents